On Friday 11 May 2007 05:16:44 am Bob Mottram wrote:
...
> But in practice it's difficult to do AI in an open source way, because
> I've found that at least up until the present there have been very few
> people who actually know anything about the algorithms involved and
> can make a useful contribution.  The typical case is that there are a
> few folks who are enthusiastic but lack either the programming ability
> or the background knowledge of AI techniques, or both.  Just learning
> about the history of AI in general so that you can recognise potential
> dead-end approaches takes some investment of time and energy.
> ...

It must be remembered that Open Source projects, at least the significant and 
successful ones, typically start around a core written by one brilliant 
individual (or very small group). Linus comes to mind, or ESR, or RMS. (Also 
think of Thomson, Kernaghan & Ritchie with C and the original Unix.) These 
were all brilliant programmers. 

http://www.catb.org/%7Eesr/writings/cathedral-bazaar/cathedral-bazaar/index.html

The innards of an OS -- scheduler, filesystem, device drivers -- are not the 
sort of thing that the average Joe Appcoder knows a lot about either. Most 
projects of that kind continue to have a small group of indispensable experts 
at their core. Given that, they can be as arcane as any project -- consider 
cryptography software -- and remember that many projects are started in 
universities as research software.

Furthermore, there are a lot of parts available as open source that can be 
included in an AI project, from numerical algorithms to theorem provers. We 
don't have to invent it, or even in many cases rewrite it. I just have to 
know why I would want to take eigenvalues or solve a SAT problem, not how to 
do it. Thus Open Source allows the AI developer to work at level well above 
the raw metal.

The 4.4bsd kernel was about 200k lines of code. Industry standard is about 20 
lines per day per programmer. If you assume that an AI is the same complexity 
as an early Unix, you're talking 27 man-years. We could imagine that one or 
two really smart people were really productive (and also got the basic design 
right!) and could do half of something that size in a few years, and the rest 
could be filled in by the community. Linux 0.001 was in Aug 1991, Linux 1.0 
in Mar 1994.

Note that a proper AI will be a learning meachine. It's quite conceivable that 
the core could be fairly small by software standards, but that the baby AI 
would need to be "raised" -- something that could be done in great profusion 
and with great variety by an O.S.-like community.

Josh







-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?member_id=231415&user_secret=fabd7936

Reply via email to